In systems where a finite resource is to be shared, a scheduler dictates how
the
resource is divided among competing processes. Examples of such systems
include,
a computer with shared CPU, a cloud computing infrastructure, a network router
etc. In such situations, the delays experienced by jobs from one user are a
function of the arrival pattern of jobs from other users, and the scheduling
policy of the server. Consequently, a scheduling system creates a timing side
channel in which information about arrival pattern from one user is
inadvertently leaked to another. In this talk, we prove that no scheduler can
provide maximum privacy without idling, and consequently no policy can
simultaneously be delay and privacy optimal.